Research On Robust Image Segmentation Algorithm Based On Neutrosophic Clustering

As the basis of image understanding and analysis,image segmentation has attracted much attention by scholars.The essence of image segmentation is the classification of pixels with similar characteristics,which means that some features are similar,such as gray,color,tonal,texture,intensity,and so on.Domestic and foreign scholars have put forward many methods of image segmentation with different principles,but there is no universal theory and method of image segmentation up to now.The quality of image segmentation plays a decisive role in the understanding and recognition of subsequent images.Explore the efficient image segmentation theory is of great importance for the development of image segmentation theory.Fuzzy set can explain the uncertainty and fuzziness of the image,which is in line with the human expression of the objective world.The clustering algorithm based on fuzzy sets have some potential advantages in image segmentation.And the fuzzy C-means clustering algorithm is one of the most widely used cluster analysis methods.The traditional FCM algorithm has some defects,such as sensitivity to initial cluster center and noise.So many improved FCM algorithms are proposed to enhance the clustering performance and noise suppression ability of FCM algorithm.However,the traditional fuzzy set only uses the membership degree to express the affirmative information that an element belongs to the set,and cannot describe the unknown uncertainty of the image processing itself.The paper introduces the neutrosophic set into fuzzy C-means clustering algorithm and proposes the neutrosophic c-means clustering algorithm.The NCM algorithm can overcome the defects of FCM algorithm,effectively cluster the class boundary and noise of class,and improve the classification accuracy,which has attracted the attention of many scholars.Although NCM algorithm has stronger ability to suppress noise than FCM algorithm,it is still difficult to meet the needs of clustering analysis of strong noise interference data samples.In order to enhance the robust performance of NCM algorithm,the theory of mahalanobis distance,hidden markov random field,hilbert kernel function and semi-supervised learning was introduced into the neutrosophic clustering algorithm.Then a series of new algorithm was proposed.The main research content are as follows:A robust color image segmentation algorithm based on neutrosophic fuzzy clustering is proposed,by introducing the Mahalanobis distance into Neutrosophic C-means Clustering algorithm.The algorithm considers the correlation between adjacent pixels of the image and the influence of the color image components.Then the spatial neighborhood information and the Mahalanobis distance are embedded in the NCM objective function.In the end,the formulas of clustering are obtained by the optimization method.In view of the deficiency of NCM algorithm,which is weak in noise resistance and difficult to cluster the non-convex and irregular data effectively,the neutrosophic fuzzy clustering segmentation algorithm based on HMRF and the semi-supervised neutrosophic clustering segmentation algorithm based on HMRF are proposed.The prior probability information of the image pixel classification is obtained by using the hidden Markov random field model.Then the information divergence is associated with the membership degree to construct the regular item,or the prior information and the membership degree are combined as the monitoring factors,which is embedded into the existing neutrosophic C-means clustering objective function.Finally the data samples in European space was mapped into Hilbert kernel space by nonlinear transformation.The new algorithm not only be able to improve the accuracy of image clustering but also improve the performance of image classification and optimize the feature of sample data in the image.In order to further improve the robustness and adaptability of NCM algorithm,a parameterized local information neutrosophic C-means clustering algorithm was proposed.The algorithm constructs the neighborhood information of pixels belonging to class,boundary and noise,respectively.Then the parameters of spatial location information with membership degree and the compactness information of neighborhood clustering are introduced to improve the anti-jamming capability of image segmentation,which make the segmentation algorithm have stronger robust performance.